PSY 9556B (Feb 5) Latent Growth Modeling

Transcription

1 PSY 9556B (Feb 5) Latent Growth Modeling Fixed and random word confusion Simplest LGM knowing how to calculate dfs How many time points needed? Power, sample size Nonlinear growth quadratic Nonlinear growth freeing loadings Piecewise models Linear growth (different ways of scaling time) Associative LGM Higher order LGM: curve-of-factors model Conditional models (time-invariant, time-variant) Multiple groups (group covariate or multiple-groups analysis) Similarity between LGM and MLM When to use LGM, when to use MLM

2 Fixed and Random word clarification Fixed and random effects in MLM and LGM Fixed effect: single values that estimate of population values (e.g., a regression coefficient, a mean intercept or slope in LGM, MLM) Random effect: provide information about the variation in the regression coefficient or intercept parameters across the clustering units (e.g., variance of intercepts and slopes in LGM or MLM) Fixed and random factors in ANOVA Fixed factor levels chosen apriori Random factor: no particular interest in the levels; chosen at random Best example of a random factor: persons Repeated measures design At least two observations nested within persons Persons as a random factor is also referred to as the clustering unit or in MLM Mixed models One fixed factor crossed with one random factor (e.g., split plot ANOVA)

3 Parameters and dfs Elements: (v (v+3))/2 = (2*5)/2 = 5 Parameters: 2 residuals (2 time points): left-over variance not explained by latent variables 1 mean intercept: the mean start-point of individual trajectories 1 mean slope: the mean slope (e.g., growth/learning/decrease) of individual trajectories 1 variance of the intercepts: variation in individual start-points 1 variance of the slopes: variation in individual slopes 1 correlation between intercept and slope (note, indicator intercepts fixed at 0) Total parameters = 7 dfs = to many parameters Number of Parameters and Degrees of Freedom Example: 2 time points (linear)

4 Parameters and dfs Elements: (v (v+3))/2 = (3*6)/2 = 9 Parameters: 3 residuals (3 time points): left-over variance not explained by latent variables 1 mean intercept: the mean start-point of individual trajectories 1 mean slope: the mean slope (e.g., growth/learning/decrease) of individual trajectories 1 variance of the intercepts: variation in individual start-points 1 variance of the slopes: variation in individual slopes 1 correlation between intercept and slope (note, indicator intercepts fixed at 0) Total parameters = 8 dfs = 9-8 = 1 Number of Parameters and Degrees of Freedom Example: 3 time points (linear)

5 Parameters and dfs Elements: (4 (4+3))/2 = (4*7)/2 = 14 Parameters: 4 residuals (4 time points): left-over variance not explained by latent variables 1 mean intercept: the mean start-point of individual trajectories 1 mean slope: the mean slope (e.g., growth/learning/decrease) of individual trajectories 1 mean quadratic component 1 variance of the intercepts: variation in individual start-points 1 variance of the slopes: variation in individual slopes 1 variance of quadratic component 3 correlations between intercept, slope, quadratic (note, indicator intercepts fixed at 0) Total parameters = 13 dfs = = 1 Number of Parameters and Degrees of Freedom Example: 4 time points (linear + quadratic)

6 Example of a LGM with Five Time Points

7 Example of a LGM with Five Time Points In this figure you should also indicate the means of the latent variables: Intercept = and Slope =

8 Example of a LGM with Five Time Points

9 Example of a LGM with Five Time Points Adding a Quadratic Component

10 Example of a LGM with Five Time Points Adding a Quadratic Component

11 Example of a LGM with Five Time Points Time Points Free

12 Example of a LGM with Five Time Points Time Points Free

13 Example of a LGM with Five Time Points Piecewise

14 Example of a LGM with Five Time Points Piecewise Poor model

15 Power in LGM Fan (2003) showed that LGM consistently showed higher statistical power for detecting group differences in the linear growth slope than repeated measures ANOVA. Study done with five time points More research needed on the impact of number of time points Fan, X. (2003). Power of latent growth modeling for detecting group differences in linear growth trajectory parameters. Structural Equation Modeling: A Multidisciplinary Journal, 10,

16 Alternative Ways of Scaling Slope: Drinking Example (8 time points)

17 Alternative Ways of Scaling Slope: Drinking Example (8 time points)

18 Associative LGM (i.e., Trajectory of Two Different Variables) I will investigate whether the slope in enjoying courses is associated with the slope in drinking. Here is a LGM of drinking only:

19 Associative LGM (i.e., Trajectory of Two Different Variables)

20 Associative LGM (i.e., Trajectory of Two Different Variables)

21 One Trajectory with Time Varying Covariates I will investigate the previous trajectory of drinking and add time varying covariates of Enjoyed University. Recall that in the previous example, Enjoyed University was modelled as a separate trajectory. In the present example, I ignore the growth factor of Enjoyed University.

22 One Trajectory with Time Varying Covariates Not a good model; covariate do not add much

23 Higher Order LGM: Curve of Factor Model

24 Higher Order LGM: Curve of Factor Model Steps: Allow residuals of indicators to correlate across time Test for strong invariance (loadings and intercepts across time points) Constrain loadings and intercepts to equality across time points Fix intercepts of the level-1 factors (i.e., univx1 univx2 in this example) at 0 In Mplus, the intercept growth factor is fixed at 0; free this parameter L-Mean Linear

25 Higher Order LGM: Curve of Factor Model

26 Higher Order LGM: Curve of Factor Model

27 LGM with Intercept and Slope Regressed on Gender (Time Invariant Covariate)

28 LGM Multiple groups (Gender)

29 LGM Multiple groups (Gender)

30 LGM Multiple groups (Gender)

31 Measures within Persons When your data file is structured in the conventional one line per subject with repeated measures on the same line

32 Measures within Persons Time (linear) Slope Drink WITHIN BETWEEN Mean = Var = Drink (intercept) Mean = Var =.782 Slope

33 Measures within Persons (Previous Model specified as LGM) Back to the conventional data structure specification IN LGM these residuals are usually not constrained to equality but they are in MLM. I constrained them here.

34 Example: Measures within Persons (LGM and MLM) LGM MLM Same as in slide 13

35 Example: Measures within Persons (LGM and MLM) LGM MLM M = Var =.0776 M = Var =

36 Example: Measures within Persons (LGM and MLM) LGM MLM